# Ideation > Generate positioned YouTube video ideas with competitive gaps, desire mapping, and shock scores. Consumes research from yt-pipeline or firecrawl-search — does NOT redo deep research. Trigger on phrases like 'video ideas for', 'what should I make a video about', 'brainstorm video ideas', 'ideation for', 'content ideas about', 'what videos should I do about X', or any request for YouTube video concept generation. - Skill: `ai-answer/ideation` (Agent Skill) - Install (CLI): `npx skillmds@latest add ai-answer/ideation` - Raw SKILL.md: https://api.skillmd.com/api/skills/ai-answer/ideation/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: AI-Answer (https://skillmd.com/u/ai-answer) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/ai-answer/ideation --- # Video Ideation Engine Generate positioned YouTube video ideas grounded in existing research and competitive analysis. Outputs structured idea briefs ready to feed into `/hooks` and `/outlines`. **This skill is strategy, not research.** It consumes research outputs from `/yt-pipeline`, `/firecrawl-search`, or vault notes. It does NOT replicate deep research — if no research exists yet, it runs a lightweight competitive scan only. ## When This Skill Activates **Activate when the user wants to:** - Brainstorm video ideas on a topic - Figure out what video to make next - Find angles on a trending topic - Generate content ideas from research they've already done **Example triggers:** - "Give me video ideas about RAG" - "What should I make a video about this week?" - "Brainstorm some Claude Code video angles" - `/ideation` or `/ideation ` ## Workflow ### Step 1 — Gather Existing Context Check what research already exists before doing anything: 1. **Search the vault** — look in `research/` and `projects/` for files related to the topic 2. **Check for recent yt-pipeline or firecrawl-search outputs** — these are your primary source material 3. **If the user references specific notes**, read them **If research exists:** Use it as the foundation. Do NOT re-research the topic. **If no research exists:** Do a lightweight competitive scan only (Step 2). Suggest the user run `/yt-pipeline ` or `/firecrawl-search ` first if they want deeper analysis. ### Step 2 — Lightweight Competitive Scan Quick YouTube search to see what exists — this is positioning intel, not deep research: > **Optional step.** This uses a separate `yt-search` helper script that does not ship with this plugin. If it isn't installed, skip the scan and generate ideas from the user's described topic and any vault/research context instead — the rest of the skill works fine without it. ```bash python "${CLAUDE_PLUGIN_ROOT}/skills/yt-search/scripts/search.py" --count 15 --months 3 ``` From results, extract: - **Saturated angles** — what's been covered to death? - **Open gaps** — what has NO or weak coverage? - **Performance outliers** — high views-to-subs ratio signals a hot angle - **Key creators** — who's covering this, how big are they? Present the landscape in 3-5 bullets. Don't dump raw search results. ### Step 3 — Generate Idea Angles Produce **7-10 video ideas** with full positioning for each: **Title (working):** A rough title — not final, just enough to capture the angle. **Angle:** What makes this different from what exists? One sentence. **Core Desire:** Map to one of the Four Horsemen: - **Money** — will this help the viewer make or save money? - **Time** — will this save them time or eliminate wasted effort? - **Health** — does this reduce stress, burnout, or cognitive load? - **Status** — does this make them look smart, capable, or ahead of the curve? **Proxy Desire (One Standard Deviation):** Don't target the core desire directly — that triggers the BS detector. Package it one step away. - Core: "Make more money" → Proxy: "Automate your entire content pipeline" - Core: "Save time" → Proxy: "Never start a Claude Code session from scratch again" - Core: "Status" → Proxy: "Build what your team thinks requires a dev shop" **Shock Score (1-100):** How surprising is this to your audience? Rate by the Law of Interestingness — distance between what the viewer believes and what this video reveals. Target 70+. **Format:** What video structure fits? - Concept/Educational ("7 RAG Concepts...") - Level-Based ("6 Levels of...") - Tool Demo ("I Built X with Y") - Contrarian Take ("X Is Ruining Your Y") - Case Study ("How I Did X") - Listicle ("5 Tools That...") - Tutorial ("How to Set Up X") - Head-to-Head ("X vs Y") **Competitive gap:** What specific gap does this fill? Reference the landscape from Step 2. ### Step 4 — Rank and Recommend Sort into three tiers: **Tier 1 — High Confidence (2-3 ideas)** Clear competitive gaps, high shock scores, proven formats. "Film tomorrow" options. **Tier 2 — Strong With Conditions (2-3 ideas)** Good angles that need research, a demo build, or timing. Worth planning for. **Tier 3 — Experimental (2-3 ideas)** Higher risk, higher ceiling. New formats, contrarian takes, untested angles. ### Step 5 — Output Structured Briefs For each Tier 1 idea, output: ``` ## [Working Title] - **Angle:** [One sentence positioning] - **Core Desire:** [Money / Time / Health / Status] - **Proxy Desire:** [One standard deviation away] - **Shock Score:** [X/100] - **Format:** [Video structure type] - **Competitive Gap:** [What's missing from existing coverage] - **Why Now:** [Timeliness signal — trending, just launched, community buzz] - **Research Depth:** [Sufficient / Thin — see Step 5b] - **Related vault notes:** [[wikilink]] [[wikilink]] ``` ### Step 5b — Deepen Research via NotebookLM (If Needed) When the user selects a Tier 1 idea, check whether the existing research actually covers that specific angle well enough to build a video. **Signs research is sufficient:** - The vault research notes directly address the chosen angle's key claims - There are specific stats, sources, or examples that support the talking points - You could write an outline right now without guessing **Signs research is thin:** - The angle is a subset or twist on the broader topic that wasn't explored in depth - Key claims in the idea brief don't have supporting data yet - You'd need to speculate or hand-wave in an outline **If research is thin and a NotebookLM notebook already exists** (from a prior `/yt-pipeline` run): 1. Find the notebook — check recent vault research files for a NotebookLM notebook ID, or list notebooks: ```bash notebooklm list --json ``` 2. Set the notebook context: ```bash notebooklm use ``` 3. Run 2-3 targeted follow-up questions against the existing sources: ```bash notebooklm ask "What specific data or examples support [the chosen angle]?" --new notebooklm ask "What counterarguments or limitations exist for [the chosen angle]?" notebooklm ask "What's the most surprising or non-obvious insight about [the chosen angle]?" ``` 4. Use the answers to refine the idea brief — update the shock score, fill in missing competitive gaps, and add specific talking points. **If no notebook exists**, suggest the user run `/yt-pipeline ` first, or proceed with lighter research via `/firecrawl-search`. **If research is sufficient**, skip this step and move to `/hooks` or `/outlines`. ### Step 6 — Save to Vault (Optional) If the user wants to save: `YOUR_VAULT_PATH/projects/YYYY-MM-DD-ideation-.md` ## Key Principles - **This is strategy, not research.** Don't replicate what `/yt-pipeline` or `/firecrawl-search` already do. Consume their outputs. - **Differentiation is everything.** An idea without a competitive gap is a wasted video. Always explain WHY this angle is different. - **Desire mapping is not optional.** Every idea maps to a Horseman + proxy desire. No map = too abstract. - **Don't pitch ideas you have already done.** Check `projects/` and `projects/archive/` before suggesting. - **Be opinionated.** "Film this one first because..." is more useful than a flat list.